Does Phase Distortion/Shift Matter in Audio? (no*)

[amirm]

/for those who still are confused - it is not any Dirac impulse, it is in fact a 96kHz sampling, reconstruction filter response to a single impulse, thus an impulse response)/

1. 96kHz 16bit 1 impulse file:

The confusion is yours. What you see above is NOT real. The audio workstation software is applying a low pass filter to create that sinc response.

2. Time domain response at DAC output (Topping D10s), sampled with Fs = 12.5 MHz:

Proving the point that you are feeding the DAC spectrum that is beyond Nyquist, causing the same as above.

3. Frequency domain response at DAC output, sampled with Fs = 1.563MHz:

That's the analog filter of the DAC as it doesn't have infinite response. If it did, it would have flat spectrum forever.

Come on folks. This is an advanced topic. If you don't know the fundamentals, please don't post.

 

[amirm]

People tend to use the term frequency response and in fact they speak about amplitude response only.

That's only half the answer. The other half is what I explained that in the case of minimum phase system, the phase is known and is not necessary to show it.

 

[Keith_W]

Come on folks. This is an advanced topic. If you don't know the fundamentals, please don't post.

On the contrary. This thread is very good for clearing up misconceptions.

 

[pma]

The other half is what I explained that in the case of minimum phase system, the phase is known and is not necessary to show it.

That's of course correct statement. The other thing is that a multi-speaker system with a complex crossover is not necessarily a minimum phase system, thus phase response cannot be simply derived from the amplitude response and thus the amplitude response (so called "frequency response" by many) is insufficient to describe the system, even if the system was a linear system.

 

[Holmz]

Sure and with it, you get high frequency spray if the previous samples are much lower value.

^Correct^.

Let’s just move away from the idea of a Dirac type of impulse.
Your position is that it is illegal.
My position is that one can easily make a file with a single nonzero sample, or they can have a bit error in a “zeros file” that then becomes the same thing.

^It^ doesn’t matter, and the main part of that post was about the step function response:

This part:

…


I think that the idea of describing how the system perturbs the phase is what is being gleaned from the step function response.

But the question for the luminaries is whether perturbations of the phase in the direct sound, off of the speaker, is important at all... And whether there is much in teh way of ABX or other studies that say it has even one half of an iota of relevance?

 

[pma]

^It^ doesn’t matter, and the main part of that post was about the step function response:

If we have a system impulse response h(t) and system step response g(t), then they are mathematically related as:

h(t) = d(g(t))/dt

Both h(t) and g(t) give a complete description of system transfer function and they carry phase information as well. Phase may be sometimes audible and sometimes not, depending on numerous factors.

 

[j_j]

I agree. And it is not very difficult to make a demonstaration.

Unfortunately you made that in the sampled domain. Yes, it does isolate the response of the DAC.

What it doesn't account for is capture. You can't get that out of the ADC unless the ADC has no antialiasing filter, in which case it's broken.

 

[pma]

Of course, I agree.

 

[amirm]

Your position is that it is illegal.
My position is that one can easily make a file with a single nonzero sample, or they can have a bit error in a “zeros file” that then becomes the same thing.

I have explained the meaning of illegal and it is not that. Just because you can digitally manipulate a file, doesn't mean what you created a valid condition for the system to respond to. Go ahead and zero out some bytes in an MPEG/JPEG video stream. You will either get a streak of bad pixels, crashed decoder, or some unexpected behavior. Video codecs have defined syntax and creating something their encoder wouldn't generate is considered an "illegal" content.

As an interesting aside, years ago someone figured out that you could cause stack overflow in audio/video/image decoders in Windows and with it, run malicious code! All you had to do is get the user to play such a file and you were golden. Clearly such content was "illegal" so the decoder had no reason to act properly but of course, must not execute such code. We fixed them but it was holding up the release of our software at the time (this is some 20 to 30 years ago).

 

[Holmz]

I have explained the meaning of illegal and it is not that. Just because you can digitally manipulate a file, doesn't mean what you created a valid condition for the system to respond to. Go ahead and zero out some bytes in an MPEG/JPEG video stream. You will either get a streak of bad pixels, crashed decoder, or some unexpected behavior. Video codecs have defined syntax and creating something their encoder wouldn't generate is considered an "illegal" content.

As an interesting aside, years ago someone figured out that you could cause stack overflow in audio/video/image decoders in Windows and with it, run malicious code! All you had to do is get the user to play such a file and you were golden. Clearly such content was "illegal" so the decoder had no reason to act properly but of course, must not execute such code. We fixed them but it was holding up the release of our software at the time (this is some 20 to 30 years ago).

We’re both triggered the “Illegal file” like ICE agents.

You are a Codex SME, so I’ll defer to you on that.
An FFT will certainly have a PSD in response to such a file.
I would think that any DAC would response to it.

But let’s just let it go.


The brunt of the that post #931 was about the whether the step function response was a worthwhile thing.
And why the Klippel supplies it?

 

[j_j]

As it happens, it is not a default graph that their software presents. You have to add a checklist to show it. So even Klippel knows it has little value.

Step response simply takes the impulse response and multiplies it by a weighting function that decreases at 6dB/octave. Nothing more. It hides a lot of the treble details. It also makes it harder to detect what's going on at high frequencies.

 

[pma]

Just because you can digitally manipulate a file, doesn't mean what you created a valid condition for the system to respond to.

It is absolutely valid to check impulse response of the DAC digital reconstruction filter + analog lowpass filter. In fact there is no other way to do it in a better way.

Edit: and it is not a "digitally manipulated file", it is a file generated in a sampled domain, see @j_j 's reply here He is the one who has got it right.

 

[amirm]

It is absolutely valid to check impulse response of the DAC digital reconstruction filter + analog lowpass filter.

And you get results that easily confuse people who don't understand that you are subjecting the system to signals that it doesn't normally see.

It is the same misguided idea of testing amplifiers with square waves created with bandwidths in Megahertz. Or how people were falling over each other in talking about "slew rate" in hi-fi amps of olden days.

Just because you can measure something, doesn't mean you should.

 

[pma]

Step response simply takes the impulse response and multiplies it by a weighting function that decreases at 6dB/octave. Nothing more. It hides a lot of the treble details. It also makes it harder to detect what's going on at high frequencies.

When we investigate transient signals in time domain (my case - measuring HV impulse waves, surge waves, chopped waves, current interruption at current zero, DC plasma arc voltage and current), we need broadband DC - MHz measuring systems. Yes, from DC. Step response is then the best test signal. If the final value ("top" of the response) has any decay with time, it tells that the system cannot transfer DC properly. And the rising edge and rise time of the step response clearly tells about high frequency -3dB corner. And we cannot allow any ringing, the response must be aperiodic without overshoot. Another requirements than in speaker audio, however still tells a lot about audio amplifiers e.g.

 

[pma]

And you get results that easily confuse people who don't understand that you are subjecting the system to signals that it doesn't normally see.

It is the same misguided idea of testing amplifiers with square waves created with bandwidths in Megahertz. Or how people were falling over each other in talking about "slew rate" in hi-fi amps of olden days.

Just because you can measure something, doesn't mean you should.

You are telling your opinions only, and your are pushing your distorted view of EE fields you are not familiar with. From your view of testing amplifiers it is clear that you have never designed any. Why do not you stick with the fields you understand.

The Scientist and Engineer's Guide to Digital Signal Processing's Table of Content

 

[Arindal]

And you get results that easily confuse people who don't understand that you are subjecting the system to signals that it doesn't normally see.

Me thinks, the same is true to all S/N measurements, THD and SINAD below a certain threshold of audibility. These signals (may it be sine waves or pink noise or jitter-induced noise) are also something our ears don't perceive, and the definition of these specs is to ensure that whatever noise is inaudible under any confusing. I find it similarly confusing to measure and publish them.

Just because you can measure something, doesn't mean you should.

Apply this to the common measurements of DACs, and you should not measure or publish anything like that, according to your logic. Have you stopped measuring such gear?

 

[amirm]

Apply this to the common measurements of DACs, and you should not measure or publish anything like that, according to your logic. Have you stopped measuring such gear?

I am applying it. I can produce a thousand tests for DACs. Indeed, AES has an entire suite and there are people online who publish books/pdfs in their measurements. I specifically avoid that and only present a curated set of tests. AP software for example can generate massive number of test results in a pdf that some other online reviewers have published. You don't see that from me.

Me thinks, the same is true to all S/N measurements, THD and SINAD below a certain threshold of audibility.

Well you think wrong. Very wrong. I don't throw impulse signal at audio electronics either for the same reason as I have explained here. People consume digital content these days with hard limits on bandwidth. It makes zero sense to test products with signals that are not in use. Doing so may force manufacturers to compromise elsewhere to meet these requirements. I have explained how this has happened in speaker design.

 

[amirm]

You are telling your opinions only, and your are pushing your distorted view of EE fields you are not familiar with. From your view of testing amplifiers it is clear that you have never designed any. Why do not you stick with the fields you understand.

The Scientist and Engineer's Guide to Digital Signal Processing's Table of Content

No, I speak as a professional who has used these technologies, not a hobbyist like you. Signal processing is one of my core expertise. I studied it in school and used it in many of the jobs I have held both in audio and video over some 50 years. I hired JJ as our chief audio architect at Microsoft. Pretty sure you would be lost after the first sentence in one of our many discussions. There is a difference between a book and having direct knowledge used in the topics we are discussing.

Anyway, the aggressive tone some of you are using is out of line. Watch your language if you want to continue to post in this thread.

 

[Holmz]

… I hired JJ as our chief audio architect at Microsoft. Pretty sure you would be lost after the first sentence in one of our many discussions. There is a difference between a book and having direct knowledge used in the topics we are discussing.

Anyway, the aggressive tone some of you are using is out of line. Watch your language if you want to continue to post in this thread.

Well no argument from me that @j_j understands the stuff, and he is concise.

Unfortunately you made that in the sampled domain. Yes, it does isolate the response of the DAC.

What it doesn't account for is capture. You can't get that out of the ADC unless the ADC has no antialiasing filter, in which case it's broken.

@j_j :
Going back to the absolutely useless topic (in the context of speaker design), can we take the Time domain impulse of the 1 sample at 48kHz and run an FFT on that?

Or push that through a DAC that band limits it below Nyquist…
And then run that into a ADC that samples at 96, 192 ?
… Would that then be 2 or 4 samples wide?

 

[Arindal]

I specifically avoid that and only present a curated set of tests

I noticed that, but I don't quite understand why you publish these test results with DACs and amps in a quite extended manner, knowing that these graphs and specs at best represent a proof of inaudibility of artifacts. And on the other hand argue that phase response, step resonse or square wave measurements in speaker reviews should not be published as they are ´confusing´ to readers.

Would rather call it particularly confusing that you are not only publishing measurements like noise spectrum graphs way below -110dB, but also sinewave graphs, THD+N figures in %, and - most confusing - compile some kind of rating based on noise measurements:

The color scheme, ranking, plus verdicts ´Excellent - very good - Fair - Poor´ intuitively indicate a difference in quality of the reviewed devices, not an equal inaudibility of the noise which you measured.

If hypothetically some lab would measure phase response or group delay of speakers, sorting them by overall phase shift from least to worst, categorizing them as ´excellent - very good - Fair - Poor´ and publishing it - wouldn't you call this ´confusing´ or ´misleading´?

I don't throw impulse signal at audio electronics either for the same reason as I have explained here.

You through pure sine waves @1K and @12K, and discuss noise/jitter artifacts -130dB below the signal. These signals are just as unrealistic and not existing from natural instruments, as impulse signals, and the noise is inaudible this or that way.

To be clear here: I am skeptical of phase response measurements being important under normal circumstances, nor do I find them intuitively understandable for laymen. Nevertheless they do show differences in a different speaker´s behavior, and there are cases in which the thresholds of audibility are most likely to being exceeded. So more relevant than noisefloor measurements with DACs and amps, in my understanding.